Scholarlink

This talk aims to examine LLMs, GenAI, and RAG, with a focus on the key challenges of mitigating hallucinations to ensure the trustworthiness and factual accuracy of their outputs. Hallucination is a well-known limitation of large language models (LLMs), reflecting their tendency to produce responses that are inaccurate, irrelevant, or inconsistent with user expectations. This undermines user confidence and makes such models less-suited for critical domains where precision and verifiability are paramount. Moreover, hallucination remains a major obstacle to achieving the levels of reliability and trustworthiness that are foundational to frameworks like the EU AI Act [1, 2], which seeks to ensure that AI systems operate safely and uphold fundamental rights. To address these issues and their broader negative impacts, both academia and industry have proposed a range of detection and mitigation strategies. The presentation will be divided into three main parts:

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

GenAI, LLM/MLLM, RAG, and Their Impacts on Hallucination, Reliability and Trustworthiness

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Hassane Essafi

摘要

This talk aims to examine LLMs, GenAI, and RAG, with a focus on the key challenges of mitigating hallucinations to ensure the trustworthiness and factual accuracy of their outputs. Hallucination is a well-known limitation of large language models (LLMs), reflecting their tendency to produce responses that are inaccurate, irrelevant, or inconsistent with user expectations. This undermines user confidence and makes such models less-suited for critical domains where precision and verifiability are paramount. Moreover, hallucination remains a major obstacle to achieving the levels of reliability and trustworthiness that are foundational to frameworks like the EU AI Act [1, 2], which seeks to ensure that AI systems operate safely and uphold fundamental rights. To address these issues and their broader negative impacts, both academia and industry have proposed a range of detection and mitigation strategies. The presentation will be divided into three main parts: